Vol. 8, No. 2, 107-113, 2009

SNNS classification of hyperspectral data of extensively used agricultural areas
Bogdan Zagajewski and Dawid Olesiuk

The goal of this paper is a presentation of artificial neural networks for land cover classifications of the DAIS 7915 hyperspectral data. The research area covers seminatural ecosystems of pastures and meadows and extensively used agricultural areas of the Low Beskid Mountains (the northern Carpathian Mts.) in southern Poland. Algorithms based on the SNNS classification (multilayer one-way perceptron with backpropagation learning method) of 3 key polygons Biesnik S, N and Wiatrowki using training sets of 60 and 40 original bands (after geometric and atmospheric correction) and first 13 MNF channels in different textural windows (1x1, 3x3, 5x5 and 7x7 pixels) were used. The results were compared with the reference sets acquired from ground validation. The best accuracy (92.1%) for the test set was achieved using 60 original bands with the 3x3 pixel subpattern size, and for the training set – 93.9% for this architecture.

View Full Text (pdf file, 5.3 MB) previous page
Submitted: 25 May 2009
Revised: 30 Sept 2009
Accepted: 01 Oct 2009
Published: 02 Oct 2009
Responsible editor: Robin Vaughan

Zagajewski B & D Olesiuk, 2009. SNNS classification of hyperspectral data of extensively used agricultural areas. EARSeL eProceedings, 8(2): 107-113


EARSeL European Association of Remote Sensing Laboratories, Strasbourg, France


BIS Library and Information System, Carl von Ossietzky University of Oldenburg


ISSN 1729-3782